Run ollama launch hermes: Ollama offers to install Hermes Agent if it is missing, lets you pick a local or cloud model, and points Hermes at the Ollama server on your Mac. Hermes Agent is an MIT licensed agent from Nous Research, and its own guides say it needs a context window of at least 64,000 tokens, which is above the 4k and 32k defaults Ollama's context page lists for under 48 GiB of VRAM, so set the context length before you judge it. Ollama's picker lists cloud models too, and those run on Ollama's service rather than your Mac. Hermes also warns that a model without tool calling can only chat, and that on CPU-only or low-VRAM setups its first reply can take minutes while the model reads Hermes's fixed prompt. This page repeats those documents and ran nothing on a Mac.
ollama launch hermes
Ollama's Hermes page says it handles four steps:
http://127.0.0.1:11434/v1 and sets your model as the primary one.To change the provider or model later, Ollama's page gives hermes model. Hermes's own setup wizard, which Ollama's page walks through for a manual install, asks for a context length and says to leave it blank to auto-detect; this page does not say what that picks for an Ollama model.
Ollama's picker offers local and cloud models together. Its cloud overview says inference for a cloud model runs on Ollama's cloud service, so those prompts leave your Mac. Does Ollama send my data? has what Ollama says about each.
Hermes's guide to running it with Ollama also carries a warning about tools: Hermes edits files, runs commands and browses through tool calls, and models without tool-call support can only chat; they cannot take actions.
Hermes's Ollama guide says Hermes requires at least 64,000 tokens for agentic work with tools, and its Mac guide calls 64K its minimum context. Ollama's context page says tasks that need a large context, such as web search, agents and coding tools, should be set to at least 64000 tokens, and warns that a larger context length increases the memory a model needs.
OLLAMA_CONTEXT_LENGTH=64000 ollama serve
That line is from Ollama's context page. The page also has an app slider and a Mac route, and the Ollama documents disagree on the default: see Ollama context length on a Mac.
ollama launch hermes-desktop prompts to install the Hermes command-line agent if it is missing, lets you pick a model, configures Hermes Desktop to use that Ollama model and opens it.hermes desktop.hermes gateway setup links Telegram, Discord, Slack, WhatsApp, Signal or Email to the agent. Hermes's security page says that with no allowlists configured and GATEWAY_ALLOW_ALL_USERS not set, all users are denied. Its gateway deployment checklist begins with setting explicit allowlists and never using GATEWAY_ALLOW_ALL_USERS=true in production.
The same page says Hermes checks each command against a list of dangerous patterns. The setting approvals.mode takes smart, manual or off, with smart the default. Its warning on off reads: "Setting approvals.mode: off disables all safety prompts. Use only in trusted environments (CI/CD, containers, etc.)."
Ollama lists both agents under one command, ollama launch, and describes them in one line each:
| Agent | Ollama's description |
|---|---|
| OpenClaw | Chat with an assistant through your messaging apps. |
| Hermes Agent | Personal assistant with memory, reusable skills, and messaging. |
Hermes's README says it can import settings, memories and skills from OpenClaw, Claude Code, Codex CLI or Cursor. hermes import-agent openclaw --dry-run previews what would be migrated, and adding --migrate-secrets also brings allowlisted API keys, which the README says happens only with that flag. For the OpenClaw side, see OpenClaw alternative for Mac.
The README gives this one-line installer for Linux, macOS and WSL2; it runs the script it downloads, so read the script first if you want to:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
| Claim | Source |
|---|---|
ollama launch hermes: install prompt, model selector, provider and endpoint, gateway | Ollama docs, Hermes Agent |
ollama launch hermes-desktop and the Desktop installer note | Ollama docs, Hermes Desktop |
Ollama lists OpenClaw and Hermes Agent under ollama launch | Ollama docs, Integrations |
| MIT licence, the installer line, the import commands | Hermes Agent README |
| The 64,000 token requirement, tool calling, the silent first reply | Hermes Agent guide: Run Hermes Locally with Ollama and Run Local LLMs on Mac |
| Apple silicon only, the DMG | Hermes Agent install page |
| Intel Macs: the command-line agent, no desktop bundle | Hermes Agent platform support |
| Local Models in Hermes Desktop | Hermes Agent docs, Local Models |
| Allowlists and command approval | Hermes Agent security page |
| At least 64000 tokens for agents and coding tools; the defaults by VRAM | Ollama docs, Context length |
| Cloud models run on Ollama's cloud service | Ollama docs, Cloud |
ollama launch hermes in a terminal, as Ollama's page describes, or open Hermes's DMG and copy Hermes.app to Applications. Hermes's install page says the macOS installer is Apple silicon only.http://127.0.0.1:11434/v1. Hermes's Mac guide also describes llama.cpp and MLX servers, and Hermes Desktop has its own Local Models route.hermes desktop.